Automatic Road Pavement Crack Detection Using a Support Vector Machine
Automatic Road Pavement Crack Detection Using a Support Vector Machine, Proc Conf. on Telecommunications - ConfTele, Castelo Baranco, Portugal, Vol. ., pp. . - ., May, 2013.
Digital Object Identifier: 0
Road surface cracks are traditionally detected by a skilled technician, while travelling along the road, for implementing a road maintenance policy. This task is very time consuming, can be dangerous and its outcomes are sometimes subjective.
A solution to automate the crack detection task by the analysis of captured road surface images is presented here. Three different pre-processing configurations are considered to smooth the texture and enhance potential cracks in images. Then, features are extracted for non-overlapping image blocks (75*75 pixels), and a support vector machine (SVM), a supervised learning algorithm, is used to classify each block as containing cracks or not. Finally a post-processing technique is applied to remove isolated crack blocks.
Two different databases are used to test the proposed algorithm. In each experiment the classifier output is compared with the respective ground-truth, provided by the manual classification of an expert. The obtained results present high recall values for both databases and also a high precision for the first database, being competitive with the best results reported in the literature.